Stelios Chatzikyriakou’s Post

I've been experimenting with gesture detection using MediaPipe and decided to open-source a small toolkit. mediapipe-gesture-signals is a lightweight Python toolkit that turns MediaPipe landmarks into stable gesture signals for real-time applications. MediaPipe is great at detecting landmarks, but most applications don't need raw coordinates every frame. They need intent signals like: touch_nose · pinch · nod · shake_head 👋 This project adds a simple layer that converts noisy landmark streams into readable gesture events that apps can actually use. The goal is simple: make gesture detection readable, reusable, and stable for interactive systems. 🔗 GitHub: https://lnkd.in/d_niqxP2 #ComputerVision #MediaPipe #AIEngineering

🚀 Released symjoy v2.6.0 I’m happy to share that I’ve released symjoy 2.6.0, a new update to my Python package for structured Unicode symbol access. symjoy provides easy access to emojis, symbols, arrows, math symbols, currency signs, and other Unicode characters through a clean and consistent API. What’s new in v2.6.0 - ⚡ Indexed search engine for faster symbol lookup - 🔎 Keyword, alias, and token indexes for metadata-driven search - 🚀 Improved search performance and registry initialization - 🧠 Better internal architecture for scalable Unicode datasets This builds on the metadata-driven architecture introduced in v2.5.0, continuing the effort to make symjoy a reliable Unicode symbol engine for Python. 📦 PyPI: https://lnkd.in/gKjtbdQ8 💻 GitHub: https://lnkd.in/gxK63QS6 Feedback and suggestions are welcome. #Python #OpenSource #SoftwareDevelopment #DeveloperTools #Unicode #GitHub #PyPI

🌟 New Blog Just Published! 🌟 📌 Boost LLM Apps with 5 Python Decorators for Speed & Savings 🚀 📖 You’re debugging a chatbot that repeats the same user query three times in a row. Each call hits the LLM API, burns $0.03, and adds a full second of latency. By the time the user sees a response, the.... 🔗 Read more: https://lnkd.in/d4jynYyF 🚀✨ #python-decorators #llm-optimization #api-cost-savings

🚀 Project #4 – Update #2: Sync Neighbors Decorator (Python Maze Generator) In this update, I worked on a problem that was bugging me in my maze generator: the neighbor updates were not always stable, which sometimes led to confused paths and inconsistent structures. So I introduced a Sync Neighbors decorator to make sure each tile and its neighbors stay properly aligned during generation. 🔧 What changed in this update: Neighbor updates are now stable and consistent The maze structure is cleaner and more reliable The result feels more playable and more structured This also makes the system easier to extend with other algorithms (DFS, Hunt & Kill, etc.) later This is part of my ongoing 100 Days Project Grind, where I’m building and improving projects step by step and sharing the progress. 🎥 Video: Project #4 – Update #2: Sync Neighbors Decorator Upgrade 💻 Code & projects: 👉 https://lnkd.in/gkGkFFJ3 More updates coming as I keep refining the system and adding new features. 🚀 Feedback and suggestions are always welcome!

Yesterday my mouse stopped working. Instead of buying a new one, I built one. I created a Virtual Mouse that runs directly from the browser. Here's how it works: • A touch zone in the browser captures finger movements. • Those movements are sent to my laptop through WebSockets. • A Python backend using PyAutoGUI interprets the coordinates. • PyAutoGUI then moves the actual system cursor in real time. So basically my phone browser became my laptop’s trackpad. Tech used: Python • PyAutoGUI • WebSockets • Browser Touch Events Sometimes the best projects come from the simplest problems. 😄 #Python #WebSockets #Automation #BuildInPublic #DeveloperLife

Count digits. Sum digits. Reverse the number. Check palindrome. Four problems, one loop. Same idea every time: n % 10 gives the last digit, n // 10 removes it. Repeat while n > 0. What you do with each digit is what changes. I wrote a step-by-step practice guide that covers: ✅ The core technique: % 10 and // 10 with a full trace (e.g. 12359) ✅ Challenge 1: Count digits (counter loop) ✅ Challenge 2: Sum of digits ✅ Challenge 3: Reverse a number (rev = rev * 10 + digit) ✅ Challenge 4: Check palindrome (save original, then compare) ✅ Pattern recognition: same loop structure, different use of the digit ✅ Common mistakes (forget n = n // 10, compare n instead of saved copy) and fixes ~9 min read. One technique, four challenges. https://lnkd.in/g4RwH_Er #Python #Programming #Coding #Beginners #LearnToCode #DigitOperations #SumOfDigits #Palindrome #Tech #SoftwareDevelopment #CodingTips

Built a content‑based movie recommendation app with Python and Streamlit using a pre‑computed movies.pkl dataset and TMDB API for posters, with duplicate ID handling and secure API key management to deliver personalized, visually rich recommendations with a responsive, dark-themed interface.

I’ve been heads down on an experimental project: a Language Model trained directly on my personal GitHub source code. It’s one thing to get a model to generate code blocks; it’s another to build the bridge that makes it usable in real time. Initially, I was sending every single character from the model’s forward pass directly over a WebSocket. While it worked, it was noisy. High frequency network operations are costly and can make it feel jittery rather than fluid. By intelligently batching decoded tokens based on both buffer length and a temporal threshold, I slashed costly WebSocket overhead. #MachineLearning #SystemDesign #SoftwareEngineering #LanguageModels #Python

We just stress-tested a FastAPI app with ~1,200 concurrent users and millions of monthly API calls. And the first thing that broke… wasn’t Python. Not even close. Real-world bottlenecks were: 1️⃣ Database connection pooling Async code was fine, bad pooling was the problem. 2️⃣ External I/O latency Storage, third-party APIs, and network latency mattered way more than CPU. 3️⃣ Zero actual observability No metrics or tracing means scaling is just a guess. The shocking truth: FastAPI never actually became a bottleneck. Our design did. After addressing the real problems: • Consistent peak traffic • Predictable latency • Simplified horizontal scaling Which leads me to wonder: How many teams are framework-optimizing… instead of addressing the actual production bottleneck? #FastAPI #Backend #Scalability #Python #SystemDesign

✅ Day 43 of 100 Days LeetCode Challenge Problem: 🔹 #684 – Redundant Connection 🔗 https://lnkd.in/gWtKJ-FA Learning Journey: 🔹 Today’s problem focused on detecting an extra edge that creates a cycle in an undirected graph. 🔹 I used the Union-Find (Disjoint Set Union) data structure to track connected components. 🔹 For each edge, I checked whether both nodes already belong to the same set. 🔹 If they do, adding that edge forms a cycle, making it the redundant connection. Concepts Used: 🔹 Union-Find (Disjoint Set) 🔹 Path Compression 🔹 Graph Cycle Detection 🔹 Connected Components Key Insight: 🔹 Union-Find provides an efficient way to detect cycles during edge additions. 🔹 Path compression optimizes repeated parent lookups. 🔹 Incrementally building connectivity helps quickly identify redundant edges. #LeetCode #DataStructures #Algorithms #CodingInterview #SoftwareEngineering #SoftwareDeveloper #ProblemSolving #Programming #ComputerScience #TechCareers #100DaysOfCode #DailyCoding #Consistency #LearningInPublic #Python #BackendDevelopment #InterviewPreparation #TechCommunity